Jev directory
CommunityJev Repos & SDKs
Jevelin
Unofficial community C#/.NET client for TypeSafe AI's Jev decision model, published on NuGet as Jevelin. It wraps the ask endpoint with typed question factories for yes/no, choice, and score questions, plus configurable model, timeout, and base address.
- Stars
- 0
- Reading time
- 2 min
- Published by
- Community
- Added
- Nov 2026
Highlights
- Install with dotnet add package Jevelin; requires the .NET 10 SDK and a TypeSafe API key.
- Supports Noul (yes/no probability), Choice (1-255 options), and Score (1-10 ordered levels) questions.
- Reads TYPESAFE_API_KEY from the environment or accepts a key in the JevClient constructor.
- State can be a string or any object that serializes to JSON, such as a record or a list of messages.
- Licensed under MIT; the demo reads the key from .NET user secrets so it never lands in the repo.
Questions about this entry
Short answers drawn from the linked project or page.
Run dotnet add package Jevelin from NuGet. It requires the .NET 10 SDK and a TypeSafe API key, which the client reads from the TYPESAFE_API_KEY environment variable or accepts via new JevClient("your-key").
No. TypeSafe publishes official SDKs for Python and JavaScript only; Jevelin is an unofficial community client covering .NET and is not affiliated with TypeSafe AI.
Three factories: JevQuestion.Noul for yes/no with a 0-1 probability, JevQuestion.Choice for picking one of 1-255 options with per-option probabilities, and JevQuestion.Score for rating on 1-10 ordered levels.
Set Model on the request, for example Model = "jev-1.13.0", for reproducible results. Otherwise the client uses DefaultModel from JevClientOptions, which defaults to jev-latest.
AskAsync throws JevApiException, which exposes StatusCode, the raw ResponseBody, and a message containing the API's detail text, so you can catch specific statuses such as Unauthorized.
Yes. JevClientOptions accepts a custom HttpClient, such as one from IHttpClientFactory. Jevelin will not dispose a client you pass in.
Image inputs and box questions, long-context mode, and listing models via GET /v1/models are listed as not supported yet. The tests use a fake HTTP handler, so they need no API key or network access.
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.